Today’s thread is how AI is reshaping everyday software work, led by the monthly What are you working on? survey, where agents showed up across developer tools, hardware projects, games, and solo workflows. That theme carries into a cautionary note from the Dark Hours apology and a more practical look at using LLMs to learn complex topics, alongside Anthropic’s feature for messaging between Claude Code sessions. Elsewhere, readers dug into software architecture and overhead, from Shopify’s Redis-to-MySQL inventory design to Windows 11’s RAM-hungry Weather app, while privacy law, email security, and a phone-as-server hack rounded out the day.
-
Ask HN: What are you working on? (August 2026)
-
news.ycombinator.com
- 571 comments
This monthly open thread asked Hacker News readers what they are building or curious about, and drew hundreds of project updates across AI tooling, developer infrastructure, games, hardware, and niche consumer apps. The standout signal was not one launch but a broad shift in how people are building: agents are now embedded in everything from CI runners and CAD-like tools to personal utilities and solo-founder workflows.
-
-
Mea Culpa – Dark Hours
-
blog.terrygodier.com
- 248 comments
The developer behind the “Dark Hours” App Store rejection story posted an apology saying AI led him to publish a web app that closely copied an existing open source astronomy app. Commenters mostly reject that explanation, pointing instead to a broader pattern of omitted facts, copied branding, and an Apple rejection story that now looks staged for attention.
-
-
Windows 11's built-in Weather app wastes more than 1 GB of RAM
-
notebookcheck.net
- 278 comments
A Notebookcheck post says Windows 11’s built-in Weather app can consume around 1 GB of RAM because it is effectively the MSN Weather site wrapped in Microsoft’s Chromium-based WebView2. The comments mostly agree the number is absurd, but add that the real issue is the browser-wrapper architecture and that Windows memory accounting makes the exact figure fuzzier than the headline suggests.
-
-
We replaced Redis with MySQL for inventory reservations and it scaled
-
shopify.engineering
- 242 comments
Shopify published an engineering post explaining how it moved inventory reservations from Redis into MySQL by using row-level locking and a bounded pool of reservable rows per item. Readers found the architecture interesting, but a big part of the conversation focused on whether the design is genuinely elegant or just a complicated way to avoid hot-row contention in checkout flash sales.
-
-
Illinois just told every operating system to start reporting your kid's age
-
itsfoss.com
- 115 comments
An Illinois bill would require operating systems and app stores to expose a user age setting or age bracket so apps can apply child-safety restrictions, and the Hacker News fight was mostly about whether that is a privacy disaster or a less-bad alternative to full ID checks. The key split was between people calling it the start of state-controlled internet and people arguing it is really standardized parental controls with no mandatory identity verification.
-
-
How I use LLMs to learn complex topics
-
laurentiugabriel.github.io
- 190 comments
A blog post shows one person’s workflow for using an LLM to generate interactive web explainers and simple simulations for unfamiliar topics, using chip manufacturing as the example. Readers found the idea creative, but most of the discussion focused on a bigger warning: AI-generated learning material is often shallow, hard to trust, and easy to mistake for real understanding.
-
-
CSS: The bomb inside your inbox
-
portswigger.net
- 33 comments
PortSwigger published a research post showing how CSS in HTML email can be abused to leak data and break isolation in major webmail clients, even when JavaScript is blocked. The comments mostly treated it as another sign that HTML email is fundamentally unsafe and that webmail sandboxes still are not strict enough.
-
-
My server is a phone now
-
seg6.space
- 234 comments
A blog post describes replacing a small personal VPS with a rooted Android phone at home, using the phone to run containers and lightweight self-hosted services. Readers mostly treated it as a fun hack with real practical caveats, especially battery safety, Android restrictions, and the fact that used mini PCs are often a simpler long-term answer.
-
-
Message your other Claude Code sessions
-
code.claude.com
- 67 comments
Anthropic documented a Claude Code feature that lets one coding-agent session message another, so parallel terminal sessions can coordinate work and share context. The comments say this pattern is already common among power users, useful for cutting token costs and context duplication, but it also expands the security and control surface in ways that feel more like RPC across trust boundaries than a harmless convenience.
-
-
Melatonin impairs morning cognition in healthy young adults (2023)
-
academic.oup.com
- 199 comments
A sleep-study abstract claiming melatonin hurts next-morning thinking in healthy young adults set off a debate mostly because the abstract omitted key details. Commenters quickly found the underlying thesis used 2 mg and 5 mg doses, which many said are far above the low doses commonly recommended for circadian timing or jet lag.
-
-
Improving Heuristics for A* Pathfinding
-
redblobgames.com
- 37 comments
A Red Blob Games article explains “differential heuristics,” a way to make A* pathfinding faster by precomputing distances to landmarks and using those to guide search. Readers mostly praised the interactive explanation, and the most useful comments focused on how the technique behaves with multiple landmarks, dynamic maps, and newer pathfinding research.
-
-
Analyzing data from Silicon Valley ventures and founders prosecuted for fraud
-
pubsonline.informs.org
- 85 comments
A research paper analyzes fraud prosecutions involving Silicon Valley startups and argues that founders often build increasingly elaborate public facades when growth stories diverge from reality. Commenters treated it less as a surprise than as a description of how venture-backed fundraising often works, especially when incentives reward hype over verification.
-
-
Incentives are for losers
-
experimental-history.com
- 117 comments
A blog post argues that treating incentives as the master explanation for human behavior becomes a moral dodge, and that people should build their own values instead of excusing bad choices as “just following incentives.” Commenters largely agreed with the critique of incentive-worship as a personal ethic, but pushed back that systems, workplaces, and public policy still have to be designed around how ordinary people actually respond to rewards, punishments, and constraints.
-
-
The Alpha 21264 CPU: NT's Greatest RISC (1998)
-
halfhill.com
- 63 comments
A 1998 BYTE article revisits DEC’s Alpha 21264, a high-end 64-bit RISC processor that ran Windows NT and briefly looked like a serious alternative to x86 workstations and servers. The comments turn it into a postmortem on why technically strong CPUs still lose, with most of the signal landing on timing, software transition pain, and market shifts that made Alpha arrive too early and die anyway.
-
-
Taxi drivers rarely die of Alzheimer's
-
theconversation.com
- 122 comments
A Conversation article argues that jobs requiring constant route planning and mental map building, especially taxi driving, may be linked to lower Alzheimer’s deaths. The comments mostly zeroed in on whether that signal is real or just a byproduct of selection effects, shorter lifespans, and overhyped science writing.
-
-
Reviving a four year old reMarkable 2
-
oskrim.github.io
- 84 comments
A blog post walks through bringing a four-year-old reMarkable 2 e-ink tablet back to life after it sat unused long enough that firmware updates stopped working, using SSH over USB and manual service tweaks. The comments turned it into a broader verdict on reMarkable: excellent hardware and unusually open Linux internals, paired with software and cloud choices that many owners still find frustrating.
-
-
Os8088: A powerful Mac-like OS for the IBM XT, 286, 386
-
os8088.com
- 159 comments
Os8088 is a hobby operating system for old IBM PC hardware that gives 8088, 286, and 386 machines a Macintosh-style graphical desktop, written largely with Claude and released with source code and a browser demo. The project impressed people as a fun proof that modern AI can produce substantial low-level software fast, but most of the conversation fixated on the site initially mislabeling it as "hand-written" and on how much credit AI-assisted work deserves.
-
-
When Feature Flags Do and Don't Make Sense (2019)
-
software.rajivprab.com
- 35 comments
A 2019 blog post argues teams overuse feature flags and should prefer canaries, testing, and fast rollbacks for most risk management. The comments mostly agree that stale flags create serious complexity, but push back hard on the idea that rollbacks can replace runtime switches in large, stateful, or multi-system deployments.
-
-
The original URL for this prediction will no longer be available in 11 years (2011)
-
longbets.org
- 95 comments
A 2011 Long Bets prediction claimed its original web address would still work 11 years later. The page is still up in 2026, but the comments focused less on the win and more on what actually keeps URLs alive, how much redirects count, and why long-term web permanence usually fails for organizational rather than technical reasons.
-
-
Making difficulty curves in games
-
davetech.co.uk
- 82 comments
A game design article lays out common ways to shape difficulty over time, from smooth ramps to spikes and dynamic adjustment. The comments mostly turned it into a practical argument about when adaptive difficulty helps players stay engaged and when it destroys the feeling of mastery or progression.
-
-
There Are Magic Hexagons of Every Order
-
gukov.dev
- 30 comments
A math blog post claims a new result about “magic hexagons”: if you relax the classic puzzle so the entries must be consecutive numbers but can start anywhere, then solutions exist for every order except 2. Readers liked the interactive explanation and focused on the new “potential field” construction, while also poking at gaps in precision and the still-pending formal proof.
-
-
Uber SubmitQueue: a high-performance speculative merge queue
-
github.com
- 57 comments
Uber open sourced SubmitQueue, the internal system it uses to speculatively rebase, test, and land multiple pull requests in parallel so a busy main branch stays usable. The comments treated it less as a novel Git feature and more as a window into the hard, expensive tooling big monorepos need once CI takes hours and commit volume gets extreme.
-
-
Honey, I shrunk the embeddings: Matryoshka vs. PCA
-
dylancastillo.co
- 19 comments
A benchmark post compares two ways to shrink embedding vectors for search systems: Matryoshka embeddings, which are trained so early dimensions carry most of the signal, and PCA, a classic dimensionality-reduction method applied after the fact. The surprising result is that plain PCA often matched or beat Matryoshka in retrieval quality at smaller vector sizes, with commenters mostly focusing on where that result is useful and where the benchmark leaves open questions.
-
-
Assert(): A Modern How To
-
fiberfs.io
- 26 comments
A blog post argues for a more deliberate, production-friendly use of `assert`, treating it as a tool for enforcing program assumptions rather than just a debug macro. The comments mostly agree with the goal but say the post blurs together assertions, input validation, exceptions, and contracts in ways that matter a lot in real systems.
-